AI summary by TwinMind
STOP Using Create React App
Theo - t3․gg
1h 19m 52s
Opening and Context Setting
- Stream triggered by Twitter drama from Swix about web ecosystem problems
- Senior engineer struggled installing json-schema-ref-parser (backend package) in frontend
- Create React App historically polyfilled Node.js modules for browser compatibility
- Goal: prevent developers from using Create React App, treat it as legacy
Stream Ground Rules
- Focus strictly on React alternatives (no Solid, Svelte, Vue, Angular)
- Respect CRA maintainers - they're working hard but fighting community expectations
- Opinion-based analysis with goal to change developer recommendations
What is Create React App
- Combination of: opinionated boilerplate, tutorial, technical implementation, build system
- Originally created when React was new and webpack configuration was difficult
- Became Facebook/Meta's blessed solution for React applications
- Contains react-scripts package with extensive custom build logic
- 323.5 MB node_modules, 300+ packages for basic setup
Problems with Create React App
Technical Issues
- No SSR (server-side rendering) story
- Ancient webpack architecture decisions
- Ejecting for customization is painful - becomes maintaining CRA fork
- Maximal approach - includes many unused features
- 300MB+ node_modules with weird polyfill behaviors
- Over 1000+ GitHub issues currently open
- Based on dated React patterns (previously included Redux, React Router)
Community Problems
- Set bad expectations about Node.js polyfills in browser
- Encouraged using CRACO (Create React App Configuration Override) - additional abstraction layer
- CRACO no longer maintained due to maintenance burden
- Bred culture of layering solutions instead of fixing root problems
- Users expect ability to npm install any Node package and have it work
Specific Technical Debt
- Previously polyfilled Node crypto, fs modules for browser use
- V5 removed many polyfills but community expects legacy behavior
- GitHub issue shows 67 downvotes on maintainer explanation of why Node polyfills are bad
- Web3/crypto packages frequently mentioned in issues - users installing backend crypto libs in frontend
Alternative Solutions Analysis
Next.js
Good:
- Meta-endorsed, well-maintained
- Modern architecture with SWC/Rust compilation
- SSR capabilities with actual HTML content from server
- Multiple rendering strategies (SSR, SSG, ISR)
- Great caching story
- Insanely fast builds
- Can replicate CRA behavior with dynamic imports + ssr: false
- Hides package complexity effectively
Bad:
- ESM story still confusing
- Unclear boundaries between frontend/backend code
- Light vendor lock-in
- GetServerSideProps largely anti-pattern
- No obvious state management solution
- Starts at 80kb gzipped
When to Use:
- HTML/SEO/Lighthouse/first paint performance matters
- Don't want to think about build problems
- Want to use current standard/what everyone's using
Vite
Good:
- Incredibly minimal and fast
- Simple templates with actual index.html
- Elegant architecture - starts from HTML, finds JS, compiles as needed
- Best for owning build system
- Great for SPAs, Electron apps, games
Bad:
- No SSR story (unsupported by creator Evan You)
- Limited to client-side applications
When to Use:
- SSR will never be concern
- Owning build system potentially valuable
- Raw speed important
- Simple requirements (Electron, games, pure SPAs)
Remix
Good:
- Built by React Router team around React Router
- Good for edge rendering
- Better React Router integration than alternatives
Bad:
- Missing HMR (hot module reload)
- Bleeding edge, less stable
- Documentation doesn't cover edge cases well
- Rough migration stories from CRA
When to Use:
- Really like React Router
- Edge rendering very important
- Bleeding edge isn't scary
- All Next.js benefits apply but prefer Router-centric approach
Migration Strategy
Decision Criteria Questions:
- How important is SSR?
- Do you need metadata/SEO?
- Do you need static renders?
- How important is hiring/team familiarity?
- Using weird client-only packages?
- How important is performance across all devices?
Migration Approach:
- If metadata/SSR important: choose Next.js or Remix
- If hiring important: choose Next.js
- If weird client packages: consider Vite
- For CRA migration to Next.js: can use dynamic imports with ssr: false to replicate CRA behavior initially
- Can redirect all routes to single page with React Router, then gradually opt into Next.js patterns
Current State Assessment
- Create React App is legacy project
- Goal should be: never start new projects with it, migrate existing ones when possible
- Next.js generally recommended for most use cases
- Vite excellent for specific SPA-only scenarios
- Remix promising but still maturing
